Exercise set

Digital Modulation and Coding Exercises

Practice digital modulation and coding problems for symbol rate, Eb/N0, MCS, OFDM, EVM, BER, HARQ, ACLR, LLR and fallback gates.

Branch: Telecommunications Engineering
Content: Exercise set
Updated: Jul 03, 2026
Revision: v1.3.7 · reviewed

These exercises practise digital modulation and coding calculations as engineering review work. They connect payload rate, coded bit rate, symbol rate, occupied bandwidth, signal-to-noise ratio, $E_b/N_0$ , modulation-and-coding selection, coding gain, OFDM overhead, spectral mask, ACLR, residual carrier-frequency offset, inter-carrier interference, phase-noise EVM, packet error, HARQ goodput, fallback service impact, interleaving latency, decoder resources, soft-decision quantization, BER confidence and validation margins.

Assume simplified screening models unless an exercise states otherwise. Real systems also require standard-specific framing, measured channel data, synchronization behavior, nonlinear distortion, phase noise, adjacent-channel interference, regulatory constraints, firmware settings, calibrated instruments and service acceptance criteria.

Release Evidence Notes

Use these exercises as screening evidence for waveform and modem-release decisions, not as proof that a selected mode will hold in service. A credible release should connect each calculation to the rate boundary, channel condition, modem configuration, impairment state, measurement setup, traffic objective and fallback action being controlled.

The minimum evidence set is:

rate evidence for payload, coded, physical-layer and application throughput boundaries, framing overhead, pilot overhead, guard intervals, retransmissions, scheduling gaps and service acceptance target;
channel evidence for received power, noise density, SINR, interference occupancy, fading state, delay spread, Doppler, phase noise, carrier-frequency offset and temperature range;
modem evidence for MCS table, coding rate, interleaver depth, decoder settings, LLR quantization, HARQ policy, synchronization lock state, equalizer state, firmware version and fallback hysteresis;
measurement evidence for instrument calibration, reference plane, bandwidth, averaging, capture duration, traffic mix, packet size, confidence level, EVM method, BER/PER counter behavior and clock stability;
validation evidence for lab-to-field correlation, packet-error target, latency budget, goodput under load, burst-error behavior, route or antenna state, alarms, rollback rule and release authority.

Treat a numerical pass as provisional when it depends on clean-channel SNR, nominal MCS thresholds, zero observed errors, average EVM, ideal coding gain, a single channel capture or a lab impairment profile that does not match the field. MCS, OFDM, HARQ, BER and LLR decisions should support release only when the evidence covers the same waveform, traffic class, receiver state and service margin.

How to Use These Exercises

For each problem:

define whether the rate is payload, coded, line, physical-layer or application throughput;
keep dB ratios separate from absolute units such as dBm;
state the modulation order, coding rate, overhead boundary and bandwidth convention;
reserve margin for implementation loss, interference, fading, temperature and measurement uncertainty;
identify the field or laboratory evidence that would validate the calculation.

The common mistake is to select a high-order modulation mode from a clean formula without checking the operating boundary. A credible release decision connects rate accounting, SNR, coding, error behavior, latency and validation evidence.

Engineering Boundary Notes

Digital modulation evidence must identify the rate boundary before any MCS decision is accepted. Payload rate, coded bit rate, symbol rate, occupied bandwidth, application goodput and fallback goodput are not interchangeable. Overheads, pilots, guard intervals, retransmissions, scheduler gaps and packetization should be visible in the same boundary as the requirement.

Link-quality evidence also has a boundary. SNR, $E_b/N_0$ , EVM, SINR, BER, PER, HARQ success and post-FEC counters answer different questions. A clean $E_b/N_0$ calculation does not prove phase-noise margin, CFO tolerance, inter-carrier interference, adjacent-channel compliance, decoder memory, latency or burst-error recovery.

Validation should match the deployed receiver state. MCS release depends on firmware version, synchronization loop settings, channel-estimation method, equalizer state, LLR quantization, confidence level, traffic mix, temperature and impairment profile. If the field channel differs from the lab model, the calculation should be treated as a hypothesis to test, not as release evidence.

Common Release Mistakes

quoting payload throughput while the calculation used coded or physical-layer rate;
selecting a high-order MCS from clean SNR while ignoring interference, phase noise, CFO or EVM reserve;
treating zero observed errors as proof without confidence level, bit count and test duration;
accepting BER while packet error, HARQ latency or backlog growth violates the service objective;
applying coding gain without checking decoder settings, interleaver depth, LLR precision and implementation loss;
passing a spectral mask without checking ACLR, power backoff and adjacent-channel coexistence;
using fallback MCS as a safety net without verifying recovery time and degraded goodput.

Scenario Map

Scenario	Main calculation	Engineering decision
Rate accounting	payload, coded bit rate, symbol rate and occupied bandwidth	Check that throughput claims use the right boundary.
Link margin	SNR, $E_b/N_0$ , EVM and release reserves	Select a mode that still works outside ideal lab conditions.
OFDM timing	cyclic prefix, pilot spacing, pilot overhead and delay-spread margin	Balance robustness, throughput, channel-estimation cost and mobility/frequency-selective fading evidence.
Receiver impairment	residual CFO, ICI, phase noise and EVM margin	Decide whether synchronization error blocks a high-order MCS.
Transmitter emission	spectral mask and ACLR guard	Decide whether waveform shaping, backoff or predistortion is needed before field release.
Error behavior	BER, PER, HARQ, interleaving, decoder latency and soft-decision quantization	Convert bit errors into service impact.
Adaptive fallback	fallback MCS, HARQ goodput, backlog growth and latency recovery	Decide whether a degraded link remains usable for the offered traffic.
Validation evidence	confidence intervals, traffic tests and release margin	Decide whether the measured evidence is strong enough for release.

Validation Package Checklist

Before treating a modulation or coding result as release evidence, collect:

required payload, coded, physical-layer and application-rate boundaries;
modulation order, coding rate, overheads, symbol rate and occupied bandwidth;
SNR, $E_b/N_0$ , SINR, EVM, BER, PER and HARQ target definitions;
channel condition, interference state, phase noise, CFO, Doppler and delay spread;
modem firmware, MCS table, synchronization, equalizer and decoder settings;
spectral mask, ACLR, backoff and regulatory or coexistence constraints;
test duration, confidence level, traffic mix, instrument calibration and uncertainty allowance;
release decision, fallback rule, rollback criterion or additional validation test.

Exercise 1: Symbol Rate and Occupied Bandwidth

A digital link must carry:

R_p=120\ \text{Mbit/s}

of payload traffic. The waveform uses 16-QAM, coding rate:

\displaystyle R_c=\frac{3}{4}

and overhead for pilots, framing and control:

\alpha_{oh}=12\%=0.12

The raised-cosine roll-off factor is:

\alpha=0.25

Find the gross coded bit rate, symbol rate, approximate occupied bandwidth and payload spectral efficiency.

Solution

16-QAM carries:

k=\log_2(16)=4\ \text{bit/symbol}

Gross coded bit rate required at the mapper boundary:

\displaystyle R_{coded}=\frac{R_p(1+\alpha_{oh})}{R_c}

\displaystyle R_{coded}=\frac{120(1.12)}{0.75}=179.2\ \text{Mbit/s}

Symbol rate:

\displaystyle R_s=\frac{R_{coded}}{k}

\displaystyle R_s=\frac{179.2}{4}=44.8\ \text{Msymbol/s}

Approximate occupied bandwidth:

B\approx R_s(1+\alpha)

B=44.8(1.25)=56.0\ \text{MHz}

Payload spectral efficiency:

\displaystyle \eta_p=\frac{R_p}{B}=\frac{120}{56.0}=2.14\ \text{bit/s/Hz}

Engineering Comment

The payload rate is not the same as the coded bit rate. Coding and overhead raise the physical-layer rate to $179.2\ \text{Mbit/s}$ before pulse shaping sets the occupied bandwidth. If an engineer used $120\ \text{Mbit/s}$ directly in the symbol-rate calculation, the bandwidth estimate would be too low and the spectrum plan would be optimistic.

Plausibility Check

An uncoded 16-QAM stream could carry $4\ \text{bit/symbol}$ , but the net result is only $2.14\ \text{bit/s/Hz}$ after coding, overhead and roll-off. That is plausible for a practical robust mode.

Exercise 2: Eb/N0 From Measured SNR

A receiver reports:

SNR=24.0\ \text{dB}

over a noise bandwidth:

B=20\ \text{MHz}

The selected modulation-and-coding mode has a coded bit rate:

R_b=54\ \text{Mbit/s}

The mode requires:

\left(E_b/N_0\right)_{req}=17.5\ \text{dB}

for the target error rate, plus:

1.5\ \text{dB}

of implementation and measurement allowance. Find the available $E_b/N_0$ margin.

Solution

Use the dB conversion:

\displaystyle \left(E_b/N_0\right)_{dB}=SNR_{dB}+10\log_{10}\left(\frac{B}{R_b}\right)

Substitute the values:

\displaystyle 10\log_{10}\left(\frac{20}{54}\right)=10\log_{10}(0.3704)=-4.31\ \text{dB}

Therefore:

E_b/N_0=24.0-4.31=19.69\ \text{dB}

Required value including allowance:

E_{req,total}=17.5+1.5=19.0\ \text{dB}

Margin:

M=19.69-19.0=0.69\ \text{dB}

Engineering Comment

The mode passes, but only by about $0.7\ \text{dB}$ . That is a thin margin for a fielded system. If interference, temperature drift, antenna misalignment, phase noise or calibration uncertainty is worse than assumed, the selected mode may become unstable.

Plausibility Check

Because the coded bit rate is higher than the noise bandwidth, $E_b/N_0$ is lower than SNR. That direction is correct: the energy assigned to each bit is spread across a high bit rate.

Exercise 3: Modulation-and-Coding Selection With Release Margin

A field link has measured:

SNR=20.5\ \text{dB}

The release rule reserves:

3.0\ \text{dB}

for fading and interference plus:

1.0\ \text{dB}

for measurement uncertainty. Candidate modes are:

Mode	Net spectral efficiency before protocol overhead	Required SNR
QPSK, rate 1/2	$1.0\ \text{bit/s/Hz}$	$4.0\ \text{dB}$
16-QAM, rate 3/4	$3.0\ \text{bit/s/Hz}$	$16.0\ \text{dB}$
64-QAM, rate 2/3	$4.0\ \text{bit/s/Hz}$	$22.0\ \text{dB}$

The channel bandwidth is:

B=10\ \text{MHz}

Protocol overhead after the physical layer is:

\alpha_{p}=10\%=0.10

Choose the highest releasable mode and estimate payload throughput.

Solution

Usable SNR after release reserves:

SNR_{usable}=20.5-3.0-1.0=16.5\ \text{dB}

Compare with the required values:

Mode	Required SNR	Passes with $16.5\ \text{dB}$ usable?
QPSK, rate 1/2	$4.0\ \text{dB}$	yes
16-QAM, rate 3/4	$16.0\ \text{dB}$	yes
64-QAM, rate 2/3	$22.0\ \text{dB}$	no

The highest releasable mode is 16-QAM, rate 3/4.

Payload throughput estimate:

\displaystyle R_p=\frac{\eta B}{1+\alpha_p}

\displaystyle R_p=\frac{3.0(10\ \text{MHz})}{1.10}=27.27\ \text{Mbit/s}

Engineering Comment

The measured SNR might seem close to 64-QAM operation if the reserves are ignored. With release reserves included, 64-QAM is clearly not acceptable. The correct decision is not the highest mode that works in a short test. It is the highest mode that still meets margin and service evidence.

Plausibility Check

The selected net spectral efficiency is $3\ \text{bit/s/Hz}$ before protocol overhead, so a $10\ \text{MHz}$ channel producing about $27\ \text{Mbit/s}$ payload is consistent.

Exercise 4: OFDM Cyclic Prefix, Overhead and Delay-Spread Margin

An OFDM waveform uses subcarrier spacing:

\Delta f=15\ \text{kHz}

The cyclic prefix duration is:

T_{cp}=4.7\ \mu\text{s}

The channel survey estimates maximum excess delay:

T_{delay}=3.2\ \mu\text{s}

and the release rule reserves:

T_{unc}=0.7\ \mu\text{s}

for measurement uncertainty and channel variation. The waveform has 600 active subcarriers, QPSK modulation and coding rate $1/2$ . Pilot and control overhead remove $12\%$ of physical-layer capacity.

Find useful symbol time, cyclic-prefix efficiency, delay-spread margin and estimated payload throughput.

Solution

Useful OFDM symbol time is approximately:

\displaystyle T_u=\frac{1}{\Delta f}

\displaystyle T_u=\frac{1}{15000}=66.67\ \mu\text{s}

Total OFDM symbol time:

T_s=T_u+T_{cp}=66.67+4.7=71.37\ \mu\text{s}

Cyclic-prefix efficiency:

\displaystyle \eta_{cp}=\frac{T_u}{T_s}=\frac{66.67}{71.37}=0.934

Delay-spread margin:

M_{cp}=T_{cp}-T_{delay}-T_{unc}

M_{cp}=4.7-3.2-0.7=0.8\ \mu\text{s}

QPSK carries:

k=2\ \text{bit/subcarrier}

Coded useful bits per OFDM symbol before pilot/control overhead:

N_{bits}=600(2)(0.5)=600\ \text{bit}

Physical-layer rate before pilot/control overhead:

\displaystyle R_{phy}=\frac{600}{71.37\times10^{-6}}=8.41\ \text{Mbit/s}

Payload rate after pilot/control overhead:

R_p=8.41(1-0.12)=7.40\ \text{Mbit/s}

Engineering Comment

The cyclic prefix passes the delay-spread screen with $0.8\ \mu\text{s}$ margin, but it also consumes time. The release decision must include both timing robustness and throughput. If the channel delay increases because of a new reflector, antenna change or industrial structure, the margin may disappear even though received power is unchanged.

Plausibility Check

The cyclic-prefix efficiency of about $93.4\%$ is plausible for this spacing and guard interval. A payload near $7.4\ \text{Mbit/s}$ is also plausible because QPSK and rate- $1/2$ coding are robust but not spectrally aggressive.

Exercise 5: Packet Error Rate From Bit Error Rate

A decoder reports post-correction bit error rate:

BER=2.0\times10^{-6}

The service transports frames of:

1500\ \text{byte}

Assume independent residual bit errors for this screening calculation. Estimate packet error rate. The service requirement is:

PER<1.0\times10^{-3}

Find the maximum acceptable residual BER for this frame size. Decide whether the mode passes.

Solution

Frame length:

N=1500(8)=12000\ \text{bit}

Packet error rate for independent residual bit errors is:

PER=1-(1-BER)^N

For small $BER$ , a useful approximation is:

PER\approx N BER

Therefore:

PER\approx 12000(2.0\times10^{-6})=0.024

The estimated packet error rate is about:

2.4\%

Maximum residual BER for the required packet error rate:

\displaystyle BER_{max}\approx \frac{PER_{req}}{N}

\displaystyle BER_{max}=\frac{1.0\times10^{-3}}{12000}=8.33\times10^{-8}

The reported residual BER:

2.0\times10^{-6}>8.33\times10^{-8}

so the mode does not pass the packet error requirement.

Engineering Comment

A residual BER that looks numerically small can still produce too many packet errors when frames are long or traffic volume is high. If the system carries retransmitted data, the service may slow down instead of visibly failing. If it carries real-time control or voice, the same residual errors may be unacceptable.

Plausibility Check

For rare independent bit errors, multiplying by frame length is a reasonable first screen. If errors are bursty, the independent model can be misleading. Interleaving, coding, packet size, fading and interference records should be reviewed before release.

Exercise 6: Interleaving Depth and Latency Tradeoff

A receiver sees burst interference events lasting:

T_b=2.0\ \text{ms}

The coded system processes one codeword every:

T_c=0.25\ \text{ms}

The decoder can correct at most 3 corrupted codeword positions within one interleaving span. A block interleaver of depth $D$ spreads a contiguous burst so that the approximate number of affected positions per decoded block is:

\displaystyle N_{aff}\approx \left\lceil \frac{N_b}{D} \right\rceil

where:

\displaystyle N_b=\frac{T_b}{T_c}

The service allows no more than:

2.0\ \text{ms}

of added interleaver latency. Choose a practical interleaver depth.

Solution

Number of codeword intervals hit by the burst:

\displaystyle N_b=\frac{2.0}{0.25}=8

The decoder condition is:

\displaystyle \left\lceil \frac{8}{D} \right\rceil \le 3

Test $D=3$ :

\displaystyle \left\lceil \frac{8}{3} \right\rceil=3

So $D=3$ is the minimum depth that meets the correction condition.

Added latency is approximately:

T_{lat}=D T_c

For $D=3$ :

T_{lat}=3(0.25)=0.75\ \text{ms}

This is below the $2.0\ \text{ms}$ service limit.

A conservative choice is $D=4$ :

\displaystyle \left\lceil \frac{8}{4} \right\rceil=2

and:

T_{lat}=4(0.25)=1.0\ \text{ms}

The practical selection is $D=4$ if memory and implementation constraints allow it.

Engineering Comment

Interleaving improves burst-error tolerance by spreading a localized disturbance across a longer coded block. It also adds latency and memory demand. The best depth is therefore not the largest possible value. It is the smallest value that gives correction margin while staying within the service latency budget.

Plausibility Check

The selected $D=4$ converts an 8-codeword burst into about 2 affected positions per decoded block, below the correction limit of 3. The added latency remains half of the service limit, leaving room for buffering, scheduling and processing variation.

Exercise 7: EVM-Based MCS Release Decision

A test receiver reports RMS error vector magnitude:

EVM=4.5\%=0.045

A common screening approximation is:

\displaystyle SNR\approx \frac{1}{EVM^2}

The release process subtracts:

$0.8\ \text{dB}$ for instrument uncertainty;
$1.2\ \text{dB}$ for implementation variation;
$1.5\ \text{dB}$ for expected interference variation.

Candidate modes require:

Mode	Required SNR before release reserves
16-QAM, rate 3/4	$16.0\ \text{dB}$
64-QAM, rate 3/4	$21.5\ \text{dB}$
256-QAM, rate 3/4	$28.0\ \text{dB}$

The policy also requires at least $2.0\ \text{dB}$ margin above the selected mode. Select the highest releasable mode.

Solution

Estimate linear SNR:

\displaystyle SNR=\frac{1}{0.045^2}=493.8

Convert to dB:

SNR_{dB}=10\log_{10}(493.8)=26.94\ \text{dB}

Subtract release reserves:

SNR_{usable}=26.94-0.8-1.2-1.5=23.44\ \text{dB}

Required SNR including the $2.0\ \text{dB}$ policy margin:

Mode	Required including policy margin
16-QAM, rate 3/4	$18.0\ \text{dB}$
64-QAM, rate 3/4	$23.5\ \text{dB}$
256-QAM, rate 3/4	$30.0\ \text{dB}$

The usable SNR is:

23.44\ \text{dB}

This is slightly below the $64$ -QAM release requirement:

23.44<23.5\ \text{dB}

Therefore the highest releasable mode is 16-QAM, rate 3/4.

Engineering Comment

The result is intentionally close. A short lab test might tempt the team to release 64-QAM, but the documented reserves make it fail by a small amount. In production engineering, a $0.06\ \text{dB}$ shortfall should not be hidden by rounding if the acceptance policy is explicit. The right response is to release the lower mode or gather stronger evidence that reduces uncertainty.

Plausibility Check

An EVM of $4.5\%$ corresponds to an idealized SNR near $27\ \text{dB}$ . After realistic reserves, the usable value is closer to $23.4\ \text{dB}$ , which is plausible for robust 16-QAM and marginal for 64-QAM.

Exercise 8: Coding Gain Versus Throughput Loss

A QPSK link is evaluated in two modes at the same symbol rate:

R_s=10\ \text{Msymbol/s}

Uncoded QPSK requires:

\left(E_b/N_0\right)_{uncoded}=9.6\ \text{dB}

for the target BER. A coded QPSK mode with code rate:

\displaystyle R_c=\frac{1}{2}

requires:

\left(E_b/N_0\right)_{coded}=5.1\ \text{dB}

but adds:

L_{impl}=0.7\ \text{dB}

of implementation loss. Protocol overhead after decoding is $8\%$ . Compute gross bit rates, net payload rate and effective coding gain.

Solution

QPSK carries:

k=2\ \text{bit/symbol}

Uncoded gross bit rate:

R_{uncoded}=R_sk=10(2)=20\ \text{Mbit/s}

Coded information bit rate before protocol overhead:

R_{coded,info}=R_skR_c=10(2)(0.5)=10\ \text{Mbit/s}

Payload after protocol overhead:

R_p=10(1-0.08)=9.2\ \text{Mbit/s}

Nominal coding gain:

G_c=9.6-5.1=4.5\ \text{dB}

Effective gain after implementation loss:

G_{eff}=4.5-0.7=3.8\ \text{dB}

Engineering Comment

Coding improves the energy margin but reduces information throughput at a fixed symbol rate. The right decision depends on service objective: weak-signal coverage, packet reliability, latency, spectrum efficiency or peak throughput. A coding gain should not be quoted without the rate and implementation boundary.

Plausibility Check

Rate- $1/2$ coding roughly halves the information rate before overhead, so a payload near $9.2\ \text{Mbit/s}$ from a $20\ \text{Mbit/s}$ QPSK symbol stream is plausible.

Exercise 9: HARQ Retransmission Goodput

Use the payload rate selected in Exercise 3:

R_p=27.27\ \text{Mbit/s}

The measured packet error rate before retransmission is:

PER=0.08

Assume independent packet errors and at most one HARQ retransmission. Estimate probability of delivery within two attempts, average transmissions per original packet, effective goodput and residual packet loss.

Solution

Probability of delivery within two attempts:

P_{succ}=1-PER^2

P_{succ}=1-0.08^2=0.9936

Average transmissions per original packet with one possible retry:

N_{tx}=1+PER=1.08

Effective goodput:

\displaystyle R_{good}=R_p\frac{P_{succ}}{N_{tx}}

\displaystyle R_{good}=27.27\frac{0.9936}{1.08}=25.1\ \text{Mbit/s}

Residual packet loss after the retry:

PER_{res}=0.08^2=0.0064

or:

0.64\%

Engineering Comment

HARQ can hide many first-pass packet errors, but it consumes airtime and adds delay. Goodput, residual loss and latency should all be reported. A high physical-layer rate with frequent retransmission may deliver less useful service than a lower MCS with fewer retries.

Plausibility Check

An $8\%$ first-pass packet error rate should reduce goodput by roughly the same order because retries consume extra transmissions. The computed drop from $27.27$ to $25.1\ \text{Mbit/s}$ is consistent.

Exercise 10: Pilot Overhead Sensitivity in an OFDM Mode

Use the physical-layer OFDM rate before pilot and control overhead from Exercise 4:

R_{phy}=8.41\ \text{Mbit/s}

The original overhead was $12\%$ . A mobility update increases pilot and control overhead to:

\alpha_{new}=20\%

Compute the new payload rate, absolute throughput loss and percentage loss relative to the original $7.40\ \text{Mbit/s}$ payload rate.

Solution

New payload rate:

R_{p,new}=R_{phy}(1-\alpha_{new})

R_{p,new}=8.41(1-0.20)=6.73\ \text{Mbit/s}

Absolute loss:

\Delta R=7.40-6.73=0.67\ \text{Mbit/s}

Percentage loss:

\displaystyle \frac{0.67}{7.40}=0.091

or:

9.1\%

Engineering Comment

Pilot overhead is not wasted if it keeps channel estimates valid, but it is still a throughput cost. Mobility, Doppler, phase noise and frequency-selective fading can force more reference symbols. The MCS decision should include this overhead rather than comparing only nominal modulation order.

Plausibility Check

Raising overhead from $12\%$ to $20\%$ removes an additional $8\%$ of the physical rate. A throughput loss near $9\%$ relative to the original payload rate is therefore reasonable.

Exercise 11: Decoder Memory and Latency Budget

A receiver uses soft-decision decoding with:

N_{cw}=4096\ \text{bit/codeword}

Soft information is stored with:

b_{LLR}=6\ \text{bit/soft\ value}

The interleaver depth selected for a robust mode is:

D=4

There are two spatial streams. Available decoder buffer memory is:

M_{max}=32\ \text{kB}

Decoder processing time is $0.18\ \text{ms}$ per codeword, interleaver latency is $1.0\ \text{ms}$ and framing latency is $0.25\ \text{ms}$ . Check memory use and latency against a $2.0\ \text{ms}$ service limit.

Solution

Buffer memory in bits:

M_{bits}=N_{cw}b_{LLR}D(2)

M_{bits}=4096(6)(4)(2)=196608\ \text{bit}

Convert to bytes:

\displaystyle M_{bytes}=\frac{196608}{8}=24576\ \text{byte}

Therefore:

M_{bytes}=24.6\ \text{kB}

Memory margin:

32-24.6=7.4\ \text{kB}

Total latency:

T_{tot}=1.0+0.18+0.25=1.43\ \text{ms}

Latency margin:

2.0-1.43=0.57\ \text{ms}

The configuration fits both memory and latency limits.

Engineering Comment

Coding and interleaving are implementation choices as well as link-budget choices. A mode that looks good in $E_b/N_0$ may fail in an embedded receiver if soft-buffer memory, decoding time, power, thermal limits or service latency are not available.

Plausibility Check

Four interleaver depths, two streams and 6-bit soft values multiply memory quickly. A result near $25\ \text{kB}$ is plausible for a modest codeword size and explains why buffer budgeting belongs in the modulation-and-coding review.

Exercise 12: Zero-Error BER Confidence Test Duration

A release test wants to demonstrate:

BER<1.0\times10^{-9}

with $95\%$ confidence using a zero-error test. For a Poisson rare-error approximation, the required number of tested bits is:

\displaystyle N\ge\frac{-\ln(1-C)}{BER_{target}}

where $C$ is confidence. The test bit rate is:

R_b=25\ \text{Mbit/s}

A proposed test duration is $60\ \text{s}$ . Determine required bits, required duration and whether the proposed test is long enough.

Solution

For $95\%$ confidence:

-\ln(1-0.95)=-\ln(0.05)=2.996

Required bits:

\displaystyle N_{req}=\frac{2.996}{1.0\times10^{-9}}=2.996\times10^9\ \text{bit}

Required duration:

\displaystyle t_{req}=\frac{2.996\times10^9}{25\times10^6}=119.8\ \text{s}

Bits in the proposed $60\ \text{s}$ test:

N_{60}=25\times10^6(60)=1.50\times10^9\ \text{bit}

Zero-error upper BER from the proposed test:

\displaystyle BER_{upper}=\frac{2.996}{1.50\times10^9}=2.0\times10^{-9}

The proposed zero-error test is not long enough to demonstrate $1.0\times10^{-9}$ at $95\%$ confidence.

Engineering Comment

Short successful tests can be statistically weak. A release report should state tested bits, confidence method, error count, traffic pattern, channel condition and whether the test covers burst errors as well as random errors. Zero observed errors is not the same as zero error probability.

Plausibility Check

At $25\ \text{Mbit/s}$ , testing about $3\times10^9$ bits should take about two minutes. A $60\ \text{s}$ test covers only half that evidence, so its confidence bound should be about twice the target BER.

Exercise 13: Residual Carrier-Frequency Offset and OFDM ICI Margin

An OFDM receiver uses subcarrier spacing:

\Delta f_{sc}=15\ \text{kHz}

After carrier recovery, the measured residual carrier-frequency offset is:

\Delta f_{CFO}=420\ \text{Hz}

The receiver has baseline RMS EVM from noise, equalization error and quantization of:

EVM_{base}=3.8\%=0.038

For a small normalized CFO, use the screening approximation:

\displaystyle \frac{P_{ICI}}{P_s}\approx\frac{\pi^2}{3}\epsilon^2

where:

\displaystyle \epsilon=\frac{\Delta f_{CFO}}{\Delta f_{sc}}

The 64-QAM release budget allows total RMS EVM no greater than:

EVM_{lim}=5.5\%

A revised carrier-recovery setting reduces residual CFO to:

\Delta f_{CFO,new}=120\ \text{Hz}

Estimate ICI power, equivalent ICI EVM, total EVM before and after retuning, and decide whether 64-QAM should be released.

Solution

Normalized CFO before retuning:

\displaystyle \epsilon=\frac{420}{15000}=0.0280

ICI power ratio:

\displaystyle \frac{P_{ICI}}{P_s}\approx\frac{\pi^2}{3}(0.0280)^2=2.58\times10^{-3}

In dB relative to the wanted subcarrier power:

10\log_{10}(2.58\times10^{-3})=-25.9\ \text{dBc}

Equivalent ICI EVM:

EVM_{ICI}=\sqrt{2.58\times10^{-3}}=0.0508=5.08\%

Combine independent EVM terms by root-sum-square:

EVM_{tot}=\sqrt{EVM_{base}^2+EVM_{ICI}^2}

EVM_{tot}=\sqrt{0.038^2+0.0508^2}=0.0634=6.34\%

Release margin before retuning:

M_{EVM}=5.5-6.34=-0.84\ \text{percentage points}

The initial condition fails the 64-QAM EVM release budget.

After retuning:

\displaystyle \epsilon_{new}=\frac{120}{15000}=0.0080

New ICI power ratio:

\displaystyle \frac{P_{ICI,new}}{P_s}\approx\frac{\pi^2}{3}(0.0080)^2=2.11\times10^{-4}

In dB:

10\log_{10}(2.11\times10^{-4})=-36.8\ \text{dBc}

New equivalent ICI EVM:

EVM_{ICI,new}=\sqrt{2.11\times10^{-4}}=0.0145=1.45\%

New total EVM:

EVM_{tot,new}=\sqrt{0.038^2+0.0145^2}=0.0407=4.07\%

New EVM margin:

M_{EVM,new}=5.5-4.07=1.43\ \text{percentage points}

ICI improvement from retuning:

-25.9-(-36.8)=10.9\ \text{dB}

The revised carrier-recovery setting makes 64-QAM releasable under this EVM budget, provided the measured PER and synchronization logs also pass.

Engineering Comment

Residual CFO in OFDM is not just a frequency-label error. It breaks subcarrier orthogonality and creates inter-carrier interference, so the impairment appears as EVM and can erase MCS margin even when received power is adequate. The release decision should use measured residual CFO, EVM, PER, carrier-loop state and channel conditions together.

Plausibility Check

The first residual CFO is only $2.8\%$ of the subcarrier spacing, yet it creates an ICI EVM term near $5\%$ under the screening model. Reducing the offset to $0.8\%$ of spacing lowers ICI power by about $11\ \text{dB}$ , which is consistent with the square-law dependence on normalized CFO.

Exercise 14: Soft-Decision LLR Quantization and Decoder Memory

A receiver stores soft-decision log-likelihood ratios, or LLRs, for a forward-error-correction decoder. The demapper clips LLRs to:

-8\le LLR\le 8

A proposed implementation uses:

b=4\ \text{bits}

per LLR value. Treat the quantizer as using $2^b$ uniformly spaced levels across the full range. The decoder team estimates that RMS LLR quantization noise should be no greater than:

\sigma_{q,allow}=0.25

in LLR units for the selected mode. Each codeword has:

N_{cw}=2048\ \text{coded bits}

The receiver stores:

N_{streams}=2

spatial streams and:

N_{HARQ}=2

HARQ processes. Compare the 4-bit and 5-bit LLR options for quantization step, RMS quantization noise, soft-buffer memory and release decision.

Solution

The full LLR range is:

R_{LLR}=8-(-8)=16

For $b$ bits, the number of quantization levels is:

N_L=2^b

Using endpoint-spaced levels, the step size is:

\displaystyle \Delta=\frac{R_{LLR}}{N_L-1}

For 4-bit LLR storage:

N_L=2^4=16

\displaystyle \Delta_4=\frac{16}{16-1}=1.067

Uniform quantization noise RMS is:

\displaystyle \sigma_q=\frac{\Delta}{\sqrt{12}}

Therefore:

\displaystyle \sigma_{q,4}=\frac{1.067}{\sqrt{12}}=0.308

The 4-bit option fails the quantization-noise allowance:

0.308>0.25

Soft-buffer memory for the 4-bit option:

B_4=N_{cw}N_{streams}N_{HARQ}b

B_4=2048(2)(2)(4)=32768\ \text{bits}

B_4=4.0\ \text{kB}

For 5-bit LLR storage:

N_L=2^5=32

\displaystyle \Delta_5=\frac{16}{32-1}=0.516

\displaystyle \sigma_{q,5}=\frac{0.516}{\sqrt{12}}=0.149

The 5-bit option passes the quantization-noise allowance:

0.149<0.25

Soft-buffer memory for the 5-bit option:

B_5=2048(2)(2)(5)=40960\ \text{bits}

B_5=5.0\ \text{kB}

Memory increase:

\Delta B=5.0-4.0=1.0\ \text{kB}

Relative memory increase:

\displaystyle \frac{\Delta B}{B_4}=\frac{1.0}{4.0}=25\%

The 5-bit option should be selected if the receiver can absorb the extra $1.0\ \text{kB}$ of soft-buffer memory and the associated bandwidth or latency cost.

Engineering Comment

Soft-decision precision is part of the link implementation, not only a firmware detail. Coarse LLR quantization can erase coding gain even when measured SNR and EVM look acceptable. The release decision should connect BER or PER evidence with LLR clipping statistics, quantizer range, soft-buffer memory, decoder latency, HARQ depth and worst-case MCS.

Plausibility Check

The 4-bit quantizer spans $16$ LLR units with only $15$ intervals, so a step slightly above $1$ LLR unit is expected. Its RMS quantization noise is about $0.31$ , above the $0.25$ allowance. Adding one bit roughly halves the step and reduces RMS noise to $0.149$ , while memory grows by $1/4$ because the stored word width rises from $4$ to $5$ bits.

Exercise 15: Phase-Noise EVM Budget for MCS Release

A receiver team wants to release a 64-QAM mode. The measured integrated RMS phase error from oscillator phase noise over the receiver’s relevant bandwidth is:

\sigma_\phi=3.2^\circ

Other measured implementation impairments, excluding phase noise, contribute RMS EVM of:

EVM_{other}=4.5\%

The 64-QAM mode has an RMS EVM limit of:

EVM_{limit}=8.0\%

The release policy reserves:

M_{policy}=1.0\ \text{percentage point}

as guard margin below the limit. Use the small-angle screen:

EVM_{\phi}\approx \sigma_{\phi,rad}

where $EVM_{\phi}$ is a linear ratio. Combine independent RMS EVM contributions by root-sum-square. Then check an improved oscillator with:

\sigma_{\phi,new}=2.2^\circ

Solution

Convert phase error to radians:

\displaystyle \sigma_{\phi,rad}=3.2\frac{\pi}{180}=0.0559\ \text{rad}

Phase-noise EVM contribution is approximately:

EVM_{\phi}=0.0559=5.59\%

Combine RMS EVM contributions:

EVM_{total}=\sqrt{EVM_{other}^2+EVM_{\phi}^2}

EVM_{total}=\sqrt{4.5^2+5.59^2}=7.17\%

The guarded EVM limit is:

EVM_{guard}=8.0-1.0=7.0\%

Guarded EVM margin is:

M_{EVM}=7.0-7.17=-0.17\ \text{percentage points}

The receiver is inside the raw $8.0\%$ EVM limit but fails the guarded release rule.

For the improved oscillator:

\displaystyle \sigma_{\phi,new,rad}=2.2\frac{\pi}{180}=0.0384\ \text{rad}

New phase-noise EVM contribution:

EVM_{\phi,new}=0.0384=3.84\%

New total RMS EVM:

EVM_{total,new}=\sqrt{4.5^2+3.84^2}=5.92\%

New guarded margin is:

M_{EVM,new}=7.0-5.92=1.08\ \text{percentage points}

The improved oscillator passes the guarded EVM release screen.

Engineering Comment

Phase noise is not interchangeable with a static carrier-frequency offset. CFO can often be estimated and corrected as a deterministic frequency error, while phase noise spreads constellation points over time and can raise EVM even when average SNR looks good. The integration bandwidth, receiver tracking loop, symbol rate, pilot structure and measurement method must therefore be stated with the phase-error number.

The root-sum-square combination is a screening model. It assumes the EVM contributors are independent and RMS-like. If phase noise, nonlinear distortion, IQ imbalance, clipping or equalizer error are correlated with the same operating condition, a measured constellation and BER/PER test under that condition should override the simplified budget.

Plausibility Check

A phase error of $3.2^\circ$ is about $0.056\ \text{rad}$ , so an EVM contribution near $5.6\%$ is expected. Combining that with an existing $4.5\%$ implementation EVM gives a total a little above $7\%$ , which explains why the raw limit passes but the guarded release fails. Reducing phase error to $2.2^\circ$ lowers the phase-noise contribution enough to recover about one percentage point of guarded EVM margin.

Exercise 16: OFDM Pilot Spacing and Coherence Margin

An OFDM modem is being released for a moving field channel. The current pilot grid has time spacing:

\Delta t_p=2.0\ \text{ms}

and frequency spacing:

\Delta f_p=180\ \text{kHz}

Channel measurements estimate coherence time:

T_c=3.2\ \text{ms}

and coherence bandwidth:

B_c=260\ \text{kHz}

The release rule requires pilot spacing no larger than half of each coherence measure:

\Delta t_p\leq0.5T_c

\Delta f_p\leq0.5B_c

The current pilot overhead is:

\alpha_{p,old}=5\%

and the physical-layer rate before pilot overhead is:

R_{phy}=80\ \text{Mbit/s}

A revised pilot grid uses:

\Delta t_{p,new}=1.0\ \text{ms}

\Delta f_{p,new}=90\ \text{kHz}

Assume pilot overhead scales inversely with the pilot spacing product:

\displaystyle \frac{\Delta t_p\Delta f_p}{\Delta t_{p,new}\Delta f_{p,new}}$$ The service requires payload throughput of at least: $$R_{req}=60\ \text{Mbit/s}$$ Check the old and new pilot grids, then decide whether the revised MCS release is defensible. ### Solution Allowed pilot time spacing: $$\Delta t_{allow}=0.5T_c=0.5(3.2)=1.6\ \text{ms}$$ Allowed pilot frequency spacing: $$\Delta f_{allow}=0.5B_c=0.5(260)=130\ \text{kHz}$$ Current time-spacing margin: $$M_t=1.6-2.0=-0.4\ \text{ms}$$ Current frequency-spacing margin: $$M_f=130-180=-50\ \text{kHz}$$ The current pilot grid fails both the time and frequency coherence screens. For the revised grid: $$M_{t,new}=1.6-1.0=0.6\ \text{ms}$$ $$M_{f,new}=130-90=40\ \text{kHz}$$ The revised grid passes both pilot-spacing screens. Pilot overhead scaling factor: $$K_p=\frac{(2.0)(180)}{(1.0)(90)}=4.0$$ New pilot overhead: $$\alpha_{p,new}=4.0(5\%)=20\%$$ Payload throughput with the old grid would be: $$R_{p,old}=80(1-0.05)=76.0\ \text{Mbit/s}$$ but the old grid is not releasable because channel estimation is undersampled. Payload throughput with the revised grid: $$R_{p,new}=80(1-0.20)=64.0\ \text{Mbit/s}$$ Throughput margin against the service requirement: $$M_R=64.0-60.0=4.0\ \text{Mbit/s}$$ The revised grid is defensible under this simplified screen: it gives enough pilot density for the measured coherence limits and still clears the payload requirement by $4.0\ \text{Mbit/s}$. ### Engineering Comment Pilot overhead is not waste when the channel is moving or frequency selective. Sparse pilots can make a high-rate mode look efficient in the rate table while the receiver is actually interpolating channel estimates beyond the coherence time or coherence bandwidth. Dense pilots reduce throughput, but they may be the only way to hold the MCS under Doppler, multipath and oscillator drift. Release evidence should include the measured Doppler or mobility basis, delay spread, coherence estimates, pilot pattern, interpolation method, equalizer logs, EVM/PER under motion and the payload boundary after pilot, control and retransmission overhead. A clean static lab channel is not enough evidence for a moving field channel. ### Plausibility Check The old grid has high payload rate because pilot overhead is only $5\%$, but both pilot spacings exceed the half-coherence rule. The revised grid quadruples pilot density and raises overhead to $20\%$, which cuts payload from $76$ to $64\ \text{Mbit/s}$. That trade is plausible: more channel-estimation evidence costs throughput but keeps the mode releasable. ## Exercise 17: ACLR Spectral-Mask Release Gate A digital transmitter is being released for an adjacent-channel coexistence requirement. The in-band channel power is measured over the assigned channel, and adjacent leakage is integrated over the adjacent channel measurement bandwidth. The release requirement is: $$ACLR_{req}=45\ \text{dB}$$ The measurement uncertainty allowance is: $$U=1.5\ \text{dB}$$ Use the guarded rule: $$ACLR_{guard}=ACLR-U$$ and require: $$ACLR_{guard}\geq ACLR_{req}$$ The initial waveform has: $$P_{ch}=18\ \text{dBm}$$ and: $$P_{adj}=-27\ \text{dBm}$$ After adding transmitter windowing and \(1\ \text{dB}\) output backoff, the measured powers are: $$P_{ch,new}=17\ \text{dBm}$$ and: $$P_{adj,new}=-33\ \text{dBm}$$ Calculate initial ACLR, guarded initial ACLR, revised ACLR, guarded revised ACLR and release decision. Also calculate the in-band power cost and adjacent-leakage improvement. ### Solution Adjacent-channel leakage ratio is: $$ACLR=P_{ch}-P_{adj}$$ Initial ACLR: $$ACLR=18-(-27)=45\ \text{dB}$$ Guarded initial ACLR: $$ACLR_{guard}=45-1.5=43.5\ \text{dB}$$ The initial waveform fails the guarded spectral-mask release gate: $$43.5<45$$ For the revised waveform: $$ACLR_{new}=17-(-33)=50\ \text{dB}$$ Guarded revised ACLR: $$ACLR_{guard,new}=50-1.5=48.5\ \text{dB}$$ The revised waveform passes: $$48.5>45$$ Guarded release margin: $$M_{ACLR}=48.5-45=3.5\ \text{dB}$$ The in-band power cost is: $$18-17=1\ \text{dB}$$ Adjacent-channel leakage improves by: $$-27-(-33)=6\ \text{dB}$$ The revised waveform trades \(1\ \text{dB}\) of in-band power for \(6\ \text{dB}\) of adjacent-channel leakage reduction and a guarded ACLR pass. ### Engineering Comment ACLR is a transmitter release metric, not only a spectrum-analyzer number. The result depends on the in-band reference power, adjacent-channel integration bandwidth, averaging, crest factor, power amplifier state, waveform windowing, digital predistortion, temperature and measurement uncertainty. A mode that barely meets the nominal mask should not be released if the guarded value fails. ### Plausibility Check The initial case has nominal \(45\ \text{dB}\) ACLR, exactly equal to the requirement, so subtracting a \(1.5\ \text{dB}\) uncertainty guard should make it fail. The revised case loses \(1\ \text{dB}\) in useful channel power but reduces adjacent leakage by \(6\ \text{dB}\), so the ACLR improves by \(5\ \text{dB}\). That explains the guarded margin of \(3.5\ \text{dB}\). ## Exercise 18: Fallback MCS Goodput and Latency Recovery Gate A digital radio link normally runs a high-rate MCS with payload capacity before HARQ airtime cost of: $$R_{normal}=72\ \text{Mbit/s}$$ The normal first-pass packet error rate is: $$PER_{normal}=0.02$$ During rain fade, the link adapts to a robust fallback MCS with payload capacity before HARQ airtime cost of: $$R_{fallback}=30\ \text{Mbit/s}$$ The fallback first-pass packet error rate is: $$PER_{fallback}=0.08$$ Assume at most one HARQ retransmission and use the screening goodput model: $$R_{good}\approx\frac{R_{mode}}{1+PER}$$ The offered service traffic is: $$R_{load}=34\ \text{Mbit/s}$$ The fade keeps the link in fallback for: $$t_f=45\ \text{s}$$ The service allows no more than: $$t_{latency,max}=150\ \text{ms}$$ of added backlog latency during degraded operation. Estimate normal goodput, fallback goodput, residual fallback packet loss after two attempts, backlog growth during the fade, peak added latency, recovery time after the link returns to normal mode, maximum fallback duration that would satisfy the latency gate, and the required traffic reduction during the 45 s fallback event. ### Solution Normal-mode goodput is: $$R_{good,normal}=\frac{72}{1+0.02}=70.59\ \text{Mbit/s}$$ Fallback goodput is: $$R_{good,fallback}=\frac{30}{1+0.08}=27.78\ \text{Mbit/s}$$ Residual packet loss after two attempts is: $$PER_{res}=PER_{fallback}^2=0.08^2=0.0064$$ So the residual fallback packet loss is: $$PER_{res}=0.64\%$$ Backlog grows during fallback because offered traffic exceeds fallback goodput: $$R_{backlog}=34-27.78=6.22\ \text{Mbit/s}$$ Peak backlog after the fade is: $$B_{peak}=R_{backlog}t_f=6.22(45)=280\ \text{Mbit}$$ In decimal megabytes: $$B_{peak}=\frac{280}{8}=35.0\ \text{MB}$$ The backlog allowed by the added-latency gate is: $$B_{allowed}=R_{load}t_{latency,max}$$ $$B_{allowed}=34(0.150)=5.10\ \text{Mbit}$$ or: $$B_{allowed}=0.6375\ \text{MB}$$ Peak added waiting time from the fallback backlog is approximately: $$t_{added}=\frac{B_{peak}}{R_{load}}$$ $$t_{added}=\frac{280}{34}=8.24\ \text{s}$$ The degraded service fails the \(150\ \text{ms}\) latency gate by a large margin. After the fade, the normal-mode excess drain rate is: $$R_{drain}=70.59-34=36.59\ \text{Mbit/s}$$ Recovery time to clear the backlog is: $$t_{recover}=\frac{280}{36.59}=7.65\ \text{s}$$ The maximum fallback duration that would satisfy the \(150\ \text{ms}\) added-latency gate is: $$t_{f,max}=\frac{B_{allowed}}{R_{backlog}}$$ $$t_{f,max}=\frac{5.10}{6.22}=0.82\ \text{s}$$ For a \(45\ \text{s}\) fallback event, the maximum admitted load is: $$R_{load,max}=R_{good,fallback}+\frac{B_{allowed}}{t_f}$$ $$R_{load,max}=27.78+\frac{5.10}{45}=27.89\ \text{Mbit/s}$$ Required traffic reduction is: $$R_{shed}=34-27.89=6.11\ \text{Mbit/s}$$ The fallback MCS should not be released for this offered traffic unless the service can shed or throttle at least about \(6.1\ \text{Mbit/s}\) during the fade, use a stronger fallback mode, reduce PER, reserve a lower offered load, or accept a different latency requirement. ### Engineering Comment Adaptive modulation and coding can keep a link connected while still breaking the service. The fallback state has acceptable physical-layer continuity, but its HARQ-adjusted goodput is below the offered traffic, so queueing becomes the dominant service failure. The release decision should therefore include degraded-mode traffic shaping, priority rules, fallback duration statistics, retransmission counters, latency probes and alarm thresholds. The normal mode recovers quickly once the fade ends, but recovery after violation is not the same as service acceptance. A protection, telemetry or voice service may fail during the \(45\ \text{s}\) degraded interval even if the backlog drains a few seconds later. ### Plausibility Check The fallback goodput is about \(28\ \text{Mbit/s}\), while offered load is \(34\ \text{Mbit/s}\), so a backlog growth near \(6\ \text{Mbit/s}\) is expected. Holding that deficit for \(45\ \text{s}\) gives hundreds of megabits of queued data, far above the \(5.1\ \text{Mbit}\) allowed by a \(150\ \text{ms}\) latency gate. The failed release decision is therefore consistent with the scale of the capacity drop. ## Review Checklist For modulation and coding exercises to become engineering decisions, verify: 1. whether every rate is payload, coded, line, physical-layer or application throughput; 2. whether SNR, $E_b/N_0$, EVM and BER are measured at the same receiver boundary; 3. whether synchronization impairments such as residual CFO, phase noise and timing jitter are included in EVM and MCS margin; 4. whether fading, interference, temperature, implementation loss and uncertainty reserves are explicit; 5. whether spectral-mask or ACLR evidence uses the correct in-band and adjacent-channel measurement bandwidths; 6. whether burst errors require interleaving, diversity, retransmission or a lower mode; 7. whether latency introduced by coding and interleaving is acceptable for the service; 8. whether HARQ, retransmission and pilot overhead are included in useful throughput; 9. whether decoder memory and soft-decision latency fit the receiver implementation; 10. whether LLR quantization range, clipping and soft-buffer width preserve coding gain; 11. whether pilot spacing is checked against coherence time, coherence bandwidth, Doppler and frequency-selective fading before claiming MCS margin; 12. whether fallback MCS goodput, retransmission cost, backlog latency and recovery time still satisfy the service during degraded intervals; 13. whether the selected mode is validated by BER, PER, FEC counters, spectrum evidence and traffic tests with enough statistical confidence. The purpose of the calculation is not to maximize the modulation order. The purpose is to release a mode that still works when the channel is no longer ideal. ## Common Mistakes - Comparing payload throughput with coded or physical-layer rate without accounting for pilots, cyclic prefix, guard time, retransmissions and protocol overhead. - Converting SNR to $E_b/N_0$ with the wrong bit rate, noise bandwidth or receiver reference point. - Selecting the highest MCS that passes nominal SNR while ignoring implementation loss, fading, interference, temperature and measurement uncertainty. - Treating EVM, BER and PER as interchangeable metrics without checking the receiver boundary, packet length, coding and test duration. - Releasing a transmitter from nominal ACLR while measurement uncertainty, adjacent-channel bandwidth or power-amplifier state is unguarded. - Assuming OFDM cyclic prefix margin is adequate from delay spread alone while residual timing error, windowing and channel-estimation error remain untested. - Increasing MCS or reducing pilots from a clean lab channel without checking coherence time, coherence bandwidth, Doppler and pilot-density evidence. - Treating carrier-frequency offset as solved after average correction while residual CFO, phase noise and inter-carrier interference still consume margin. - Using a short zero-error BER run as proof of very low BER without computing the statistical confidence bound. - Increasing interleaving depth to hide burst errors without checking latency, memory, HARQ timing and service requirements. - Crediting coding gain without including code-rate loss, decoder complexity, soft-decision precision and LLR clipping. - Treating adaptive fallback as acceptable because the link stays up while degraded goodput, backlog latency and traffic shedding are untested. - Passing a modem mode in a lab channel and releasing it without field evidence from fading, interference, mobility, temperature, oscillator tolerance and traffic load.

REF